A method for identifying a region in a space and related apparatus

Abstract

The application discloses a region identification method and related device in space, applied to the field of artificial intelligence, and the method comprises the steps that a unit division is performed on a space material image of a to-be-identified space in a bird's-eye view to obtain a to-be-identified unit in the to-be-identified space and a first material image of the to-be-identified unit; a second material image of the to-be-identified unit in an object view is obtained; key point clustering is performed on key point information in object activity information of the to-be-identified unit according to a preset time period to obtain a first time sequence of the to-be-identified unit; first, second and third deep features are obtained by inputting the to-be-identified unit into a feature extractor in a region identification model respectively; and a first fusion feature obtained through feature fusion is input into a classification layer in the region identification model to identify the region of the to-be-identified unit, so that the region identification result of the to-be-identified unit is obtained. Even if there are different regions with similar material forms in the space, the region identification error can be reduced, and the high-precision requirement of region identification is met.

Description

Technical Field

[0001] This application relates to the field of search technology, and in particular to a method and related apparatus for identifying regions in space. Background Technology

[0002] With the gradual planning of different areas in space, timely and accurate identification of these areas is of great significance for spatial optimization and management. Currently, with the rapid development of artificial intelligence technology, spatial area identification methods typically employ AI technology; that is, AI-based area identification models are applied to spatial area identification methods.

[0003] In related technologies, spatial region recognition methods rely on spatial material images. Specifically, the method involves first extracting features from image blocks segmented from the spatial material image to obtain image features, and then using a region recognition model to perform region recognition on the image features to obtain the region recognition result.

[0004] However, spatial material images do not provide a comprehensive description of regions in space. When there are different regions in space that are similar in material form, relying solely on spatial material images to identify regions in space using the above method can easily lead to region identification errors and make it difficult to meet the high-precision requirements for spatial region identification. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method and related apparatus for identifying regions in space. Even when different regions in space have similar material forms, it can reduce the occurrence of region identification errors and improve the accuracy of region identification results, thereby meeting the high-precision requirements for region identification in space.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] On one hand, embodiments of this application provide a method for identifying regions in space, the method comprising:

[0008] The spatial material image of the space to be identified, viewed from an overhead perspective, is divided into units to obtain the unit to be identified in the space and the first material image of the unit to be identified; the space to be identified contains multiple objects.

[0009] Acquire a second material image of the unit to be identified from the object's perspective;

[0010] Based on a preset time period, key point clustering is performed on the key point information in the object activity information of the unit to be identified to obtain the first time series of the unit to be identified.

[0011] The feature extractor in the region recognition model is used to extract features from the first material image, the second material image and the first time series respectively to obtain the first deep feature, the second deep feature and the third deep feature.

[0012] The region identification result of the unit to be identified is obtained by performing region identification on the first fused feature obtained by feature fusion of the first depth feature, the second depth feature and the third depth feature through the classification layer in the region identification model.

[0013] On the other hand, embodiments of this application provide a spatial region identification device, the device comprising: a segmentation unit, an acquisition unit, a clustering unit, an extraction unit, and an identification unit;

[0014] The segmentation unit is used to divide the spatial material image of the space to be identified into units from an overhead view, thereby obtaining the unit to be identified in the space to be identified and the first material image of the unit to be identified; the space to be identified contains multiple objects.

[0015] The acquisition unit is used to acquire a second material image of the unit to be identified from the perspective of the object;

[0016] The clustering unit is used to perform key point clustering on the key point information in the object activity information of the unit to be identified according to a preset time period, so as to obtain the first time series of the unit to be identified.

[0017] The extraction unit is used to extract features from the first material image, the second material image, and the first time series respectively through the feature extractor in the region recognition model to obtain a first depth feature, a second depth feature, and a third depth feature.

[0018] The recognition unit is used to perform region recognition on the first fused feature obtained by feature fusion of the first depth feature, the second depth feature and the third depth feature through the classification layer in the region recognition model, and obtain the region recognition result of the unit to be recognized.

[0019] On the other hand, embodiments of this application provide a computer device, the computer device including a processor and a memory:

[0020] The memory is used to store computer programs and to transfer the computer programs to the processor;

[0021] The processor is configured to execute the method described in any of the foregoing aspects according to instructions in the computer program.

[0022] On the other hand, embodiments of this application provide a computer-readable storage medium for storing a computer program for performing the methods described in any of the foregoing aspects.

[0023] On the other hand, embodiments of this application provide a computer program product, including a computer program that, when run on a computer device, causes the computer device to execute the method described in any of the foregoing aspects.

[0024] As can be seen from the above technical solution, firstly, the spatial material image of the space to be identified, which contains multiple objects, is divided into units from an overhead view to obtain the unit to be identified and the first material image of the unit to be identified; second material image of the unit to be identified from the object's perspective is obtained; and key point clustering is performed on the key point information in the object activity information of the unit to be identified according to a preset time period to obtain the first time series of the unit to be identified; based on the description of the material information of the unit to be identified from the overhead view in the first material image and the supplementary description of the material information of the unit to be identified from the object's perspective in the second material image, the first time series is added to further describe the object activity pattern of the unit to be identified, so as to further reflect the regional characteristics of the area to which the unit to be identified belongs. Then, the first material image, the second material image, and the first time series are input into the feature extractor of the region recognition model for feature extraction, resulting in first, second, and third depth features. The first fused feature, obtained by fusing the first, second, and third depth features, is then input into the classification layer of the region recognition model for region recognition, yielding the region recognition result for the unit to be identified. The first fused feature integrates the material features of the unit to be identified from a bird's-eye view, the material features from an object's perspective, and object activity features reflecting the regional characteristics of the region to which the unit belongs. This provides a more comprehensive description of the regional features of the unit to be identified, making the region recognition results more accurate. This method reduces region recognition errors and improves the accuracy of region recognition results even when different regions with similar material morphologies exist in space, thus meeting the high-precision requirements for region recognition in space. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1A schematic diagram of a system architecture for a spatial region identification method provided in an embodiment of this application;

[0027] Figure 2 A flowchart illustrating a spatial region identification method provided in this application embodiment;

[0028] Figure 3 A schematic diagram illustrating the process stages of a spatial region identification method provided in an embodiment of this application;

[0029] Figure 4 A schematic diagram of the model structure of a region identification model provided in an embodiment of this application;

[0030] Figure 5 A schematic diagram of a portion of sample units in a sample space provided in an embodiment of this application;

[0031] Figure 6 This is a schematic diagram illustrating the confidence level of a region identification model in identifying the correct region to which a sample unit belongs, provided in an embodiment of this application.

[0032] Figure 7 This application provides a schematic diagram illustrating the confidence distribution of a region identification model that correctly identifies the region to which a sample unit belongs, as shown in an embodiment of the present application.

[0033] Figure 8 A flowchart illustrating a method for identifying target functional areas in urban space, provided as an embodiment of this application;

[0034] Figure 9 A structural diagram of a spatial region identification device provided in an embodiment of this application;

[0035] Figure 10 A structural diagram of a server provided in an embodiment of this application;

[0036] Figure 11 This is a structural diagram of a terminal provided in an embodiment of this application. Detailed Implementation

[0037] The embodiments of this application will now be described with reference to the accompanying drawings.

[0038] Currently, with the gradual planning of different areas within a space, timely and accurate identification of these areas is crucial for spatial optimization and management. For example, with rapid economic and social development and the needs of urban spatial planning, the scope of urban space is gradually expanding. Different functional areas within the city are converging and merging, blurring and disappearing the boundaries between them, creating a unique phenomenon of coexistence and integration. For instance, target functional areas within an urban space are informal residential areas characterized by low spatial quality, chaotic land use, high building density, lack of infrastructure, and poor environmental quality, negatively impacting the urban space. Therefore, timely and accurate identification of target functional areas within an urban space is essential for urban spatial optimization and management.

[0039] In related technologies, region recognition models based on artificial intelligence are applied to spatial region recognition methods. These methods rely on spatial material images. Specifically, they involve first extracting features from image blocks segmented from the spatial material image to obtain image features, and then using a region recognition model to perform region recognition on these image features to obtain the region recognition result. For example, first, feature extraction is performed on image blocks segmented from a material image of urban space to obtain image features, and then a target functional region recognition model is used to perform target functional region recognition on these image features to obtain the target functional region recognition result.

[0040] However, research has revealed that spatial material images are not comprehensive enough in describing spatial regions. When different regions in a space have similar material forms, relying solely on spatial material images for region identification using the aforementioned method is prone to errors and fails to meet the high-precision requirements for spatial region identification. For example, in urban spaces where there are target functional areas and old residential areas with similar material forms, relying solely on urban space material images for target functional area identification is likely to result in errors and fails to meet the high-precision requirements for target functional area identification in urban spaces.

[0041] To address the aforementioned technical problems, this application provides a method for identifying regions in space. The method involves dividing a spatial material image of a space containing multiple objects from an overhead view into units, obtaining the unit to be identified and a first material image of the unit; acquiring a second material image of the unit from an object's perspective; and performing key point clustering on key point information in the object activity information of the unit to be identified according to a preset time period, obtaining a first time series of the unit to be identified. Based on the first material image describing the material information of the unit to be identified from an overhead view, and the second material image supplementing the description of the material information from an object's perspective, the first time series is added to further describe the object activity patterns of the unit to be identified, thereby further reflecting the regional characteristics of the area to which the unit to be identified belongs. The first material image, the second material image, and the first time series are input into the feature extractor of the region recognition model for feature extraction, resulting in first, second, and third depth features. These three depth features are then fused to obtain a first fused feature, which is input into the classification layer of the region recognition model for region recognition, yielding the region recognition result for the unit to be identified. The first fused feature incorporates material features from a bird's-eye view, material features from an object's perspective, and object activity features reflecting the regional characteristics of the region to which the unit belongs. This provides a more comprehensive description of the regional features of the unit to be identified, making the region recognition results more accurate. This method reduces region recognition errors and improves the accuracy of region recognition results, even when different regions with similar material morphologies exist in space, thus meeting the high-precision requirements for region recognition in space.

[0042] Next, the system architecture of the spatial region identification method will be introduced. See [link / reference] Figure 1 , Figure 1 This is a schematic diagram of a system architecture for a spatial region identification method provided in an embodiment of this application. The system architecture includes a terminal 101 and a server 102.

[0043] The terminal 101 includes, but is not limited to, smartphones, tablets, laptops, desktop computers, smart voice interaction devices, smart home appliances, and in-vehicle terminals. The server 102 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal 101 and server 102 can be directly or indirectly connected via wired or wireless communication, and this application does not impose any restrictions. For example, the terminal 101 and server 102 can be connected via a network, which can be wired or wireless.

[0044] Terminal 101 acquires a spatial material image of the space to be identified from an overhead view and sends the spatial material image to server 102. Server 102 divides the spatial material image of the space to be identified from an overhead view into units, obtaining the unit to be identified and the first material image of the unit to be identified in the space. The space to be identified contains multiple objects. As an example, the space to be identified is "City A", the spatial material image of "City A" from an overhead view is "Remote Sensing Image A", and the objects are "vehicles". Terminal 101 can acquire "Remote Sensing Image A" of "City A" with multiple "vehicles" gathered from an overhead view and send "Remote Sensing Image A" to server 102. Server 102 can divide "Remote Sensing Image A" into units, obtaining the unit to be identified in "City A" as "Unit 1" and the first material image of "Unit 1" as "Remote Sensing Image Block 1".

[0045] Server 102 acquires a second material image of the unit to be identified from the perspective of the object. As an example, based on the above example, server 102 can also acquire a second material image of "unit 1" from the perspective of the "vehicle" as "street view image 21", "street view image 22", ... and "street view image 2m", where m is a positive integer and m > 2.

[0046] Server 102 performs key point clustering on key point information in the object activity information of the unit to be identified according to a preset time period to obtain the first time series of the unit to be identified. As an example, the preset time period is "every 1 hour", the object activity information is "vehicle travel information", the key point information is "travel origin information" and "travel destination information", and the key point clustering is "travel origin clustering" and "travel destination clustering". Based on the above example, server 102 can also perform "travel origin clustering" and "travel destination clustering" on the "travel origin information" and "travel destination information" in "vehicle travel information" of "unit 1" according to "every 1 hour" to obtain "travel origin time series 1" and "travel destination time series 1" of "unit 1", and concatenate "travel origin time series 1" and "travel destination time series 1" of "unit 1" to obtain the first time series of "unit 1" as "time series 1".

[0047] Server 102 extracts features from the first material image, the second material image, and the first time series using the feature extractor in the region recognition model, obtaining first depth features, second depth features, and third depth features. As an example, the region recognition model is a "target functional area recognition model." Based on the above example, server 102 can input "remote sensing image patch 1," "street view image 21," "street view image 22," ..., "street view image 2m," and "time series 1" into the feature extractor in the "target functional area recognition model" for feature extraction, obtaining the first depth feature as "depth feature 1," the second depth feature as "depth feature 2," and the third depth feature as "depth feature 3."

[0048] Server 102 uses the classification layer in the region recognition model to perform region recognition on a first fused feature obtained by feature fusion of the first deep feature, the second deep feature, and the third deep feature, to obtain the region recognition result of the unit to be identified. As an example, based on the above example, server 102 can fuse "deep feature 1", "deep feature 2", and "deep feature 3" to obtain a first fused feature as "fused feature 1", and input "fused feature 1" into the classification layer in the "target functional region recognition model" to perform target functional region recognition, thereby obtaining the target functional region recognition result of "unit 1".

[0049] In other words, the spatial material image of the space to be identified, which contains multiple objects, is divided into units from an overhead view to obtain the unit to be identified and the first material image of the unit; a second material image of the unit to be identified is obtained from the object's perspective; and key point clustering is performed on the key point information in the object activity information of the unit to be identified according to a preset time period to obtain the first time series of the unit to be identified; based on the description of the material information of the unit to be identified from the overhead view in the first material image and the supplementary description of the material information of the unit to be identified from the object's perspective in the second material image, the first time series is added to further describe the object activity pattern of the unit to be identified, so as to further reflect the regional characteristics of the area to which the unit to be identified belongs. The first material image, the second material image, and the first time series are input into the feature extractor of the region recognition model for feature extraction, resulting in first, second, and third depth features. These three depth features are then fused to obtain a first fused feature, which is input into the classification layer of the region recognition model for region recognition, yielding the region recognition result for the unit to be identified. The first fused feature incorporates material features from a bird's-eye view, material features from an object's perspective, and object activity features reflecting the regional characteristics of the region to which the unit belongs. This provides a more comprehensive description of the regional features of the unit to be identified, making the region recognition results more accurate. This method reduces region recognition errors and improves the accuracy of region recognition results, even when different regions with similar material morphologies exist in space, thus meeting the high-precision requirements for region recognition in space.

[0050] Based on the above examples, this method, building upon the description of the material information of "Unit 1" in "City A" from an overhead perspective by "Remote Sensing Image A," and the supplementary description of the material information of "Unit 1" from a "vehicle" perspective by "Street View Image 21," "Street View Image 22," ..., and "Street View Image 2m," by adding "Time Series 1" to further describe the "vehicle" activity patterns of "Unit 1," thereby further reflecting the regional characteristics of the area to which "Unit 1" belongs. Furthermore, "Fusion Feature 1," after feature extraction and fusion, integrates the material features of "Unit 1" from the overhead perspective, the material features of "Unit 1" from the "vehicle" perspective, and "vehicle travel features" reflecting the regional characteristics of the area to which "Unit 1" belongs. This provides a more comprehensive description of the regional characteristics of "Unit 1." Based on "Fusion Feature 1," the regional identification of the target functional area of ​​"Unit 1" becomes more accurate. Even when there are target functional areas and old residential areas with similar material morphology in "City A," this method can reduce the occurrence of errors in target functional area identification and improve the accuracy of the identification results, thus meeting the high-precision requirements for target functional area identification in urban spaces.

[0051] The spatial region identification method provided in this application relates to Artificial Intelligence (AI) technology. AI refers to the theories, methods, technologies, and application systems that utilize digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine capable of reacting in a manner similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making capabilities.

[0052] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics.

[0053] The spatial region recognition method provided in this application mainly involves artificial intelligence technologies, including Computer Vision (CV), Natural Language Processing (NLP), and Machine Learning (ML) / Deep Learning. Computer vision is a science that studies how to enable machines to "see." More specifically, it refers to using cameras and computers to replace human eyes in recognizing, tracking, and measuring targets, and further processing the images to create images more suitable for human observation or transmission to instruments. As a scientific discipline, computer vision researches related theories and technologies, attempting to establish artificial intelligence systems capable of extracting information from images or multidimensional data. Computer vision technologies typically include image processing, image recognition, and image semantic understanding.

[0054] Natural Language Processing (NLP) is an important field within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language people use in daily life—and thus it has a close relationship with linguistic research. NLP techniques typically include text processing and text semantic understanding.

[0055] Machine learning is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory, among others. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instruction-based learning.

[0056] In specific embodiments of this application, if the object activity information and other related data involve user information, when the above embodiments of this application are applied to specific products or technologies, it is necessary to obtain separate permission or consent from the user, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0057] It should be noted that, in the embodiments of this application, the computer device can be a server or a terminal, and the method provided in the embodiments of this application can be executed by the terminal or the server alone, or by the terminal and the server in cooperation. Figure 1 The corresponding embodiments are mainly described using the server executing the method provided in the embodiments of this application as an example.

[0058] Furthermore, when the method provided in this application embodiment is executed by the terminal alone, its execution method is the same as... Figure 1 The corresponding embodiments are similar, mainly replacing the server with a terminal. Furthermore, when the terminal and server cooperate to execute the methods provided in this application's embodiments, steps that need to be displayed on the front-end interface can be executed by the terminal, while steps that require background calculations and do not need to be displayed on the front-end interface can be executed by the server.

[0059] Next, taking the method provided in the embodiments of this application as an example, and in conjunction with the accompanying drawings, the spatial region identification method provided in the embodiments of this application will be described in detail. See also Figure 2 , Figure 2 A flowchart of a spatial region identification method provided in this application embodiment, the method including:

[0060] S201: Divide the spatial material image of the space to be identified into units from an overhead view to obtain the unit to be identified and the first material image of the unit to be identified in the space to be identified; the space to be identified contains multiple objects.

[0061] In this embodiment, the space requiring region identification is designated as the space to be identified. This space contains multiple objects. To identify regions within the space promptly and accurately, it is necessary to acquire material images that describe the material information of the space. Considering that describing the material information of the space from different perspectives provides a more comprehensive picture, and that describing the spatial material information from an overhead view offers advantages such as a wide viewing angle, rich material information, and convenient acquisition, it is first necessary to acquire spatial material images that describe the spatial material information of the space from an overhead view—that is, to acquire spatial material images of the space to be identified from an overhead view. In practical applications, spatial material images of the space to be identified from an overhead view can be, for example, remote sensing images or UAV images.

[0062] Based on this, considering the distribution of different regions in the space to be identified, the subsequent process is not to directly identify the region of the space to be identified, but to divide the spatial material image into units to obtain the units to be identified in the space to be identified and the first material image of the units to be identified, so as to identify the region to which the units to be identified belong in a timely and accurate manner. The units to be identified can be, for example, the grid to be identified in the space to be identified obtained by dividing the spatial material image into grids, and the first material image is used to describe the material information of the units to be identified from an overhead view.

[0063] As an example, when the spatial region recognition method is applied to the target functional area recognition scenario in urban space, the space to be identified is "City A", the object is "vehicles", and the spatial material image of "City A" with multiple "vehicles" gathered in the overhead view is "remote sensing image A". The server divides "City A" into units in the overhead view of "remote sensing image A", and the unit to be identified in "City A" is "unit 1" and the first material image of "unit 1" is "remote sensing image block 1".

[0064] S202: Acquire a second material image of the unit to be identified from the perspective of the object.

[0065] In this embodiment, after executing S201 to obtain the unit to be identified in the space to be identified and the first material image of the unit to be identified, since the first material image is used to describe the material information of the unit to be identified from a bird's-eye view, in order to more comprehensively describe the material information of the unit to be identified, considering that describing the local material information of the unit to be identified from the object's perspective has advantages such as detailed material information and expressing the visible environment and information behind the environment, it is also necessary to obtain a second material image for describing the local material information of the unit to be identified from the object's perspective; that is, to obtain a second material image of the unit to be identified from the object's perspective. In practical applications, the second material image of the unit to be identified from the object's perspective can be, for example, a street view image or a social media image.

[0066] In the specific implementation of S202, multiple local material images used to describe the local material information of the space to be identified from the object's perspective can be obtained first; that is, multiple local material images of the space to be identified from the object's perspective can be obtained. Then, the multiple local material images are connected according to the unit to be identified to determine whether the multiple local material images have a connection relationship with the unit to be identified. The local material images that have a connection relationship with the unit to be identified are used as the second material images of the unit to be identified from the object's perspective. Therefore, this application provides a possible implementation method. S202 can be, for example, as follows: connecting multiple local material images of the space to be identified from the object's perspective according to the unit to be identified to obtain the second material image.

[0067] Among them, the multiple local material images of the space to be identified from the perspective of the object are collected from the sampling points in the space to be identified from the perspective of the object. The sampling points are unevenly distributed in the space to be identified, so the number of local material images that are connected to the unit to be identified is uncertain. That is, the number of local material images that are connected to different units to be identified may be the same or different.

[0068] As an example, based on the above example, the multiple local material images of "City A" from the perspective of "vehicles" are "street view image 21", "street view image 22", ... and "street view image 2n", where n is a positive integer and n > 3; the server performs unit concatenation on "street view image 21", "street view image 22", ... and "street view image 2n" according to "unit 1" to obtain the second material image of "unit 1" from the perspective of "vehicles" as "street view image 21", "street view image 22", ... and "street view image 2m", where m is a positive integer and 2 < m < n.

[0069] Of course, multiple local material images of the unit to be identified from the object's perspective can also be directly acquired as the second material image of the unit to be identified from the object's perspective. These multiple local material images of the unit to be identified from the object's perspective are obtained from sampling points within the unit to be identified from the object's perspective. Therefore, this application provides a possible implementation method, and S202 can specifically be, for example, directly acquiring multiple local material images of the unit to be identified from the object's perspective and determining them as the second material image.

[0070] As another example, based on the above example, the server directly obtains multiple local material images of "Unit 1" from the perspective of "vehicle" as "street view image 21", "street view image 22", ... and "street view image 2m", that is, the second material image of "Unit 1" from the perspective of "vehicle" is "street view image 21", "street view image 22", ... and "street view image 2m"; where m is a positive integer, m>2.

[0071] S203: Perform key point clustering on the key point information in the object activity information of the unit to be identified according to the preset time period to obtain the first time series of the unit to be identified.

[0072] Since spatial material images do not provide a comprehensive description of regions in space, when different regions in space have similar material morphologies, relying solely on spatial material images for region identification in related technologies can easily lead to region identification errors and fail to meet the high-precision requirements of spatial region identification.

[0073] Therefore, in this embodiment of the application, in order to more comprehensively describe the regional information of the area to which the unit to be identified belongs, and considering the aggregation of key point information in the object activity information of the unit to be identified within a preset time period, it is necessary to dynamically describe the object activity pattern of the unit to be identified, so as to further reflect the regional characteristics of the area to which the unit to be identified belongs. This requires obtaining object activity information for dynamically describing the object activity pattern of the unit to be identified, and performing key point clustering on the key point information in the object activity information of the unit to be identified according to a preset time period to obtain a first time series of the unit to be identified. This allows for more accurate identification of the area to which the unit to be identified belongs based on the first time series. The first time series is used to dynamically describe the object activity pattern of the unit to be identified, so as to further reflect the regional characteristics of the area to which the unit to be identified belongs.

[0074] In practical applications, the activity information of the unit to be identified can be the vehicle travel information, vehicle check-in information, or vehicle communication information, etc.

[0075] In the specific implementation of S203, either of the following two implementation methods can be adopted:

[0076] One specific implementation of S203 refers to: after obtaining object activity information used to dynamically describe the activity patterns of objects in the space to be identified (i.e., based on obtaining object activity information of the space to be identified), key point information of the space to be identified can be obtained through key point extraction, then key point information of the units to be identified can be obtained through unit connection, and finally, the first time series of the units to be identified can be obtained through key point clustering. Specifically, firstly, key points are extracted from the object activity information of the space to be identified to obtain key point information of the space to be identified; then, unit connection is performed on the key point information of the space to be identified according to the units to be identified to determine whether there is a connection relationship between the key point information of the space to be identified and the units to be identified, and the key point information that has a connection relationship with the units to be identified is used as the key point information of the units to be identified; finally, key point clustering is performed on the key point information of the units to be identified according to a preset time period to obtain the first time series of the units to be identified. Therefore, this application provides a possible implementation, and S203 may include, for example, the following S2031-S2033:

[0077] S2031: Extract key points from the object activity information in the space to be identified to obtain key point information of the space to be identified.

[0078] S2032: Connect the cells according to the key point information of the space to be identified, and obtain the key point information of the cells to be identified.

[0079] S2033: Perform key point clustering on the key point information of the unit to be identified according to the preset time period to obtain the first time series.

[0080] The above implementation of S2031-S2033 involves first extracting key point information of the space to be identified based on the object activity information of the space to be identified, and then obtaining key point information of the unit to be identified through unit connection. The amount of information data of the key point information of the space to be identified is much smaller than the amount of information data of the object activity information of the space to be identified, making the subsequent unit connection process simpler and more convenient.

[0081] As an example, the preset time period is "every 1 hour", the object activity information is "vehicle travel information", the key point extraction is "travel origin extraction" and "travel destination extraction", and the key point clustering is "travel origin clustering" and "travel destination clustering". Based on the above example, the server performs "travel origin extraction" and "travel destination extraction" on the "vehicle travel information" of "City A" to obtain the "travel origin information" and "travel destination information" of "City A"; according to "Unit 1", the "travel origin information" and "travel destination information" of "City A" are processed. The "trip origin information" and "trip destination information" are connected in separate units to obtain the "trip origin information" and "trip destination information" of "Unit 1". The "trip origin information" and "trip destination information" of "Unit 1" are clustered into "trip origin clustering" and "trip destination clustering" respectively according to "every 1 hour" to obtain the "trip origin time series 1" and "trip destination time series 1" of "Unit 1". The "trip origin time series 1" and "trip destination time series 1" of "Unit 1" are then concatenated to obtain the first time series of "Unit 1" as "time series 1".

[0082] Another specific implementation of S203 refers to: after obtaining object activity information used to dynamically describe the object activity patterns of the space to be identified (i.e., based on obtaining the object activity information of the space to be identified), the object activity information of the unit to be identified can be obtained through unit connection, then the key point information of the unit to be identified can be obtained through key point extraction, and finally the first time series of the unit to be identified can be obtained through key point clustering. Specifically, firstly, unit connection is performed on the object activity information of the space to be identified according to the unit to be identified to determine whether there is a connection relationship between the object activity information of the space to be identified and the unit to be identified. The object activity information that has a connection relationship with the unit to be identified is used as the object activity information of the unit to be identified. Then, key points are extracted from the object activity information of the unit to be identified to obtain the key point information of the unit to be identified. Finally, key point clustering is performed on the key point information of the unit to be identified according to a preset time period to obtain the first time series of the unit to be identified. Therefore, this application provides a possible implementation, and S203 may include, for example, the following S2034-S2036:

[0083] S2034: Connect the units according to the object activity information of the space to be identified, and obtain the object activity information of the unit to be identified.

[0084] S2035: Extract key points from the object activity information of the unit to be identified to obtain key point information of the unit to be identified.

[0085] S2036: Perform key point clustering on the key point information of the unit to be identified according to a preset time period to obtain the first time series.

[0086] The above implementation of S2034-S2036 involves first obtaining the object activity information of the unit to be identified through unit connection based on the object activity information of the space to be identified, and then obtaining the key point information of the unit to be identified through key point extraction. The amount of information data of the object activity information of the unit to be identified is much smaller than the amount of information data of the object activity information of the space to be identified, making the subsequent key point extraction process simpler and more convenient.

[0087] As an example, based on the above example, the server performs unit connection on the "vehicle travel information" of "City A" according to "Unit 1" to obtain the "vehicle travel information" of "Unit 1"; "travel origin extraction" and "travel destination extraction" are performed on the "vehicle travel information" of "Unit 1" to obtain the "travel origin information" and "travel destination information" of "Unit 1"; "travel origin clustering" and "travel destination clustering" are performed on the "travel origin information" and "travel destination information" of "Unit 1" according to "every 1 hour" to obtain the "travel origin time series 1" and "travel destination time series 1" of "Unit 1", and the "travel origin time series 1" and "travel destination time series 1" of "Unit 1" are concatenated to obtain the first time series of "Unit 1" as "Time Series 1".

[0088] Furthermore, in this embodiment, to reduce the adverse effects of the presence of singular values ​​in the first time series of the unit to be identified on subsequent region identification, the first time series of the unit to be identified can be normalized to obtain a normalized first time series. This normalized first time series is then used to update the first time series of the unit to be identified, thereby reducing the likelihood of singular values ​​in the updated first time series and thus mitigating the aforementioned adverse effects. Therefore, this application provides a possible implementation method, which may further include the following S1-S2:

[0089] S1: Normalize the first time series to obtain the normalized first time series.

[0090] S2: Update the first time series based on the normalized first time series.

[0091] Furthermore, in this embodiment, the execution order of S202 and S203 is not limited. S202 can be executed first and then S203, or S203 can be executed first and then S202, or S202 and S203 can be executed simultaneously.

[0092] S204: Using the feature extractor in the region recognition model, feature extraction is performed on the first material image, the second material image, and the first time series to obtain the first depth feature, the second depth feature, and the third depth feature.

[0093] In this embodiment, after executing S201-S203 to obtain the first material image, the second material image, and the first time series of the unit to be identified in the space to be identified, the first material image is used to describe the material information of the unit to be identified from an overhead view, the second material image is used to describe the material information of the unit to be identified from an object view, and the first time series is used to dynamically describe the object activity pattern of the unit to be identified, so as to further reflect the regional characteristics of the area to which the unit to be identified belongs. The combination of the first material image, the second material image, and the first time series can more comprehensively describe the regional information of the area to which the unit to be identified belongs. It is also necessary to input the first material image, the second material image, and the first time series into the feature extractor in the region recognition model for feature extraction to obtain the first depth feature corresponding to the first material image, the second depth feature corresponding to the second material image, and the third depth feature corresponding to the first time series, so that the area to which the unit to be identified belongs can be more accurately identified based on the first depth feature, the second depth feature, and the third depth feature.

[0094] Among them, the first depth feature represents the material characteristics of the unit to be identified from an overhead view, the second depth feature represents the material characteristics of the unit to be identified from an object view, and the third depth feature represents the object activity characteristics that reflect the regional characteristics of the area to which the unit to be identified belongs.

[0095] As an example, the region identification model is called the "Target Functional Region Identification Model". Based on the above example, the server inputs "Remote Sensing Image Patch 1", "Street View Image 21", "Street View Image 22", ..., "Street View Image 2m" and "Time Series 1" into the feature extractor in the "Target Functional Region Identification Model" for feature extraction, and obtains the first depth feature as "Depth Feature 1", the second depth feature as "Depth Feature 2" and the third depth feature as "Depth Feature 3".

[0096] In the specific implementation of S204, since the first material image, the second material image, and the first time series of the unit to be identified in the space to be identified are input data from three different data sources, based on this, the feature extractor in the region identification model is built with three different types of first convolutional neural networks, convolutional-recurrent neural networks, and recurrent-fully convolutional neural networks, which are used to extract features from the input data from the aforementioned three different data sources respectively, to obtain first deep features, second deep features, and third deep features.

[0097] In this system, a first convolutional neural network is used to extract features from a first material image, obtaining a high-dimensional feature vector as a first deep feature; a convolutional-recurrent neural network is used to extract features from a second material image, obtaining a high-dimensional feature vector as a second deep feature; and a recurrent-fully convolutional neural network is used to extract features from a first time series, obtaining a high-dimensional feature vector as a third deep feature. Therefore, this application provides a possible implementation where the feature extractor in the region recognition model includes a first convolutional neural network, a convolutional-recurrent neural network, and a recurrent-fully convolutional neural network; S204 may include, for example, the following S2041-S2043:

[0098] S2041: The first material image is used to extract features through the first convolutional neural network in the region recognition model to obtain the first depth features.

[0099] As an example, considering limitations in data volume and computational resources, the first convolutional neural network could be, for instance, the classic, relatively low-complexity 18-layer residual network (ResNet18) in computer vision. The network parameters of ResNet18 are trained and fine-tuned based on ImageNet, a large visualization database used for research in visual object recognition software. Alternatively, the first convolutional neural network could be other convolutional neural networks; for example, given sufficient data and computational resources, it could be a 101-layer residual network (ResNet101), etc.

[0100] S2042: The second material image is used to extract features by using a convolutional-recurrent neural network in the region recognition model to obtain the second depth features.

[0101] In the specific implementation of S2042, since the number of second material images of the units to be identified in the identification space is uncertain, a convolutional-recurrent neural network (CRN) combining a second convolutional neural network and a first recurrent neural network is used to handle the problem of uncertain number during feature extraction. Based on this, each image in the second material image is set as a time step. In each time step, the second convolutional neural network is used to extract features from the image corresponding to that time step in the second material image, obtaining a high-dimensional feature vector as the first convolutional feature corresponding to that time step. The first recurrent neural network is used to calculate and process the first convolutional feature to obtain the output feature corresponding to that time step. Finally, the output feature obtained from the second convolutional neural network and the first recurrent neural network in the last time step is used as the second depth feature. Therefore, this application provides a possible implementation method, where the convolutional-recurrent neural network includes a second convolutional neural network and a first recurrent neural network; S2042 may include, for example, the following S3-S4:

[0102] S3: The second convolutional neural network in the region recognition model is used to extract features from the second material image to obtain the first convolutional features.

[0103] S4: The first convolutional features are processed by the first recurrent neural network in the region recognition model to obtain the second deep features.

[0104] As an example, the convolutional recurrent neural network can be, for example, a Vision-Long Short Term Memory (Vision-LSTM) network; the second convolutional neural network can be, for example, the classic and relatively low-complexity ResNet18 in computer vision, whose network parameters are trained and fine-tuned on the Places365 dataset containing 1,800,000 images with 365 landscape classifications; in addition, the second convolutional neural network can also be other convolutional neural networks, such as ResNet101, etc., when there is sufficient data and computing resources; the first recurrent neural network can be, for example, a Long Short Term Memory (LSTM) network, and in addition, the first recurrent neural network can also be other recurrent neural networks, such as a Gated Recurrent Unit (GRU) or a bidirectional recurrent neural network (RNN), etc.

[0105] S2043: By using a recurrent fully convolutional neural network in the region recognition model, features are extracted from the first time series to obtain the third deep features.

[0106] In a specific implementation of S2043, the recurrent-fully convolutional neural network (RCN) is obtained by combining a second recurrent neural network (RNN) and a fully convolutional neural network (WCNN). The second RNN is used to extract features from the first time series, obtaining a high-dimensional feature vector as the second convolutional feature; the WCNN is used to extract features from the first time series, obtaining a high-dimensional feature vector as the temporal feature. Based on this, the second convolutional feature and the temporal feature are fused to obtain the third deep feature. Therefore, this application provides a possible implementation method where the RNN includes a second RNN and a fully convolutional neural network; S2043 may include, for example, the following S5-S7:

[0107] S5: The second recurrent neural network in the region recognition model is used to extract features from the first time series to obtain the second convolutional features.

[0108] S6: Use the fully convolutional neural network in the region recognition model to extract features from the first time series and obtain time features.

[0109] S7: The second convolutional feature and the temporal feature are fused to obtain the third deep feature.

[0110] The fusion process can be, for example, a splicing process. Of course, in this embodiment, the specific implementation of the fusion process is not limited, as long as the fusion of the second convolutional features and the temporal features is achieved.

[0111] As an example, a convolutional recurrent neural network can be, for example, a Long Short Term Memory-Fully Convolutional Network (LSTM-FCN); the second recurrent neural network can be, for example, an LSTM, or other recurrent neural networks, such as a GRU or a bidirectional RNN; all layers in a fully convolutional neural network (FCN) are convolutional layers.

[0112] S205: Through the classification layer in the region recognition model, the first fused feature obtained by feature fusion of the first deep feature, the second deep feature and the third deep feature is used to perform region recognition, and the region recognition result of the unit to be identified is obtained.

[0113] In this embodiment, after obtaining the first depth feature, second depth feature, and third depth feature through S204, since the first depth feature represents the material features of the unit to be identified from a bird's-eye view, the second depth feature represents the material features of the unit to be identified from an object's perspective, and the third depth feature represents the object activity features reflecting the regional characteristics of the area to which the unit to be identified belongs, it is necessary to fuse the first depth feature, the second depth feature, and the third depth feature to obtain a first fused feature that fuses the material features of the unit to be identified from a bird's-eye view, the material features of the unit to be identified from an object's perspective, and the object activity features reflecting the regional characteristics of the area to which the unit to be identified belongs. The first fused feature is then input into the classification layer in the region identification model for region identification, thereby predicting the region to which the unit to be identified belongs as the region identification result of the unit to be identified.

[0114] As an example, based on the above example, the server fuses "Deep Feature 1", "Deep Feature 2" and "Deep Feature 3" to obtain the first fused feature as "Fused Feature 1", and inputs "Fused Feature 1" into the classification layer of the "Target Functional Region Recognition Model" to identify the target functional region, and obtains the target functional region recognition result of "Unit 1".

[0115] Based on the above descriptions in S201-S205, see [link to relevant documentation]. Figure 3 , Figure 3 This is a schematic diagram of the process stages of a spatial region identification method provided in an embodiment of this application; wherein, the above-mentioned S201-S205 can be divided into two process stages; the first process stage may be to perform data preprocessing on the spatial material image of the space to be identified from an overhead view, multiple local material images of the space to be identified from an object view, and object activity information of the space to be identified, respectively, to obtain the first material image of the unit to be identified, the second material image of the unit to be identified, and the first time series of the unit to be identified; the second process stage may be to input the first material image, the second material image, and the first time series into the region identification model for region identification, to obtain the region identification result of the unit to be identified.

[0116] The first process stage specifically refers to: dividing the spatial material image of the space to be identified from an overhead view into units to obtain the unit to be identified and the first material image of the unit to be identified; connecting multiple local material images of the space to be identified from the object's perspective based on the unit to be identified to obtain the second material image of the unit to be identified; extracting key points of the object activity information in the space to be identified to obtain key point information of the space to be identified; connecting the key point information of the space to be identified based on the key point information of the unit to be identified to obtain the key point information of the unit to be identified; and clustering the key point information of the unit to be identified according to a preset time period to obtain the first time series.

[0117] See Figure 4 , Figure 4 This is a schematic diagram of the model structure of a region recognition model provided in an embodiment of this application. The region recognition model includes a feature extractor and a classification layer. The feature extractor includes a first convolutional neural network, a convolutional-recurrent neural network, and a recurrent-fully convolutional neural network. Based on this, the second process stage refers to: inputting a first material image of the unit to be identified into the first convolutional neural network for feature extraction to obtain a first deep feature; inputting a second material image of the unit to be identified into the convolutional-recurrent neural network for feature extraction to obtain a second deep feature; inputting a second time series of the unit to be identified into the recurrent-fully convolutional neural network for feature extraction to obtain a third deep feature; and inputting a first fused feature obtained by feature fusion of the first, second, and third deep features into the classification layer for region recognition to obtain the region recognition result of the unit to be identified.

[0118] Furthermore, in this embodiment, the aforementioned region recognition model is obtained by training a preset recognition model based on the third material image of the sample unit from an overhead view, the fourth material image of the sample unit from an object view, the object activity information of the sample unit, and the region label data identifying the region to which the sample unit belongs. The specific training process involves: first, obtaining training samples for training the preset recognition model to obtain the region recognition model; dividing the spatial material image of the sample space from an overhead view into units to obtain sample units and their third material images; obtaining the fourth material image of the sample unit from an object view; and performing keypoint clustering on the keypoint information in the object activity information of the sample unit according to a preset time period to obtain the second time series of the sample unit; based on this, the third material image, the fourth material image, and the second time series of the sample unit constitute the training samples. The sample unit has region label data identifying the region to which it belongs.

[0119] Then, considering the step-by-step recognition method of first extracting features and then identifying regions based on a preset recognition model, which involves using the third material image, the fourth material image, and the second time series, it is difficult to train the parameters of the feature extraction process when training the model parameters of the preset recognition model. This leads to the problem of insufficient feature extraction accuracy in the aforementioned step-by-step recognition method. Therefore, to solve this problem, the feature extraction process and the region recognition process are integrated into the preset recognition model. That is, the preset recognition model includes a feature extractor and a classification layer. First, the third material image, the fourth material image, and the second time series are respectively input into the feature extractor in the preset recognition model for feature extraction, obtaining the fourth depth feature representing the material characteristics of the sample unit from a bird's-eye view, the fifth depth feature representing the material characteristics of the sample unit from an object's view, and the sixth depth feature representing the object activity characteristics of the region to which the sample unit belongs. Then, the second fused feature obtained by feature fusion of the fourth, fifth, and sixth depth features is input into the classification layer in the preset recognition model for region recognition, obtaining the region prediction result of the sample unit.

[0120] Finally, based on the region label data and region prediction results of the sample units, the loss is calculated using the loss function of the preset recognition model, and the model parameters of the preset recognition model are trained. That is, the model parameters of the feature extractor and classification layer in the preset recognition model are trained until the preset recognition model converges or reaches the preset number of iterations to complete training. The trained preset recognition model is then used as the region recognition model. Therefore, this application provides a possible implementation method, and the training steps of the first target matching model may include, for example, the following S8-S13:

[0121] S8: Divide the spatial material image of the sample space into units from an overhead view to obtain sample units and third material images of the sample units; each sample unit has regional label data that identifies the region to which it belongs.

[0122] As an example, the sample space is "City B," which contains multiple "vehicles." The spatial image of "City B" from an overhead view is "Remote Sensing Image B," which is actually a remote sensing image covering the entire city of "City B," with a resolution of 0.6 meters and including red, green, and blue channels. The target functional area label data in "Remote Sensing Image B" is annotated by referring to remote sensing images, street view images, and planning documents. Sample units in "City B" are obtained by dividing the city into 500m × 500m units. Sample units belonging to the target functional area are considered positive samples, while those not belonging to the target functional area are considered negative samples. Furthermore, to verify the effectiveness of the area recognition model, 20% of the sample units are randomly selected as the test set, and the remaining 80% are used as the training set. See [link to relevant documentation]. Figure 5 , Figure 5 This is a schematic diagram of a portion of sample units in a sample space provided in an embodiment of this application; wherein, the portion of sample units includes positive samples in the training set, negative samples in the training set, positive samples in the test set, and negative samples in the test set.

[0123] S9: Obtain the fourth material image of the sample unit from the object's perspective.

[0124] As an example, based on the above example, the fourth material image of the sample unit in "City B" from the perspective of "vehicle" can be a "street view image" collected from "X Map". It is obtained by accessing the application interaction interface of sampling points from the perspective of "vehicle" every 50 meters along the road network in "City B". Each sampling point from the perspective of "vehicle" collects images from four directions: front, back, left and right, as well as the coordinate position of the sampling point.

[0125] S10: Perform key point clustering on the object activity information of the sample unit according to the preset time period to obtain the second time series of the sample unit.

[0126] As an example, based on the above example, the object activity information of the sample unit in "City B" is "vehicle travel information". Specifically, "vehicle travel information" can be collected by sensors such as smart locators on the vehicle, including vehicle number, vehicle location, vehicle travel time, vehicle origin information and vehicle destination information.

[0127] S11: Using the feature extractor in the preset recognition model, feature extraction is performed on the third material image, the fourth material image, and the second time series to obtain the fourth deep feature, the fifth deep feature, and the sixth deep feature.

[0128] S12: By using the classification layer in the preset recognition model, the second fusion feature obtained by feature fusion of the fourth, fifth and sixth deep features is used to identify the region and obtain the region prediction result of the sample unit.

[0129] S13: Based on the regional label data, regional prediction results, and the loss function of the preset recognition model, train the model parameters of the preset recognition model to obtain the regional recognition model.

[0130] The number of sample units in different regions may vary significantly, exceeding a preset difference value. To prevent the trained region recognition model from heavily favoring regions with a large number of sample units, a weighted cross-entropy loss function is required during training. The weights in this loss function are used to balance the difference in the number of sample units. Therefore, this application provides a possible implementation where, when the difference in the number of sample units in different regions exceeds a preset difference value, the loss function is specifically a weighted cross-entropy loss function, with the weights used to balance the difference in the number of sample units.

[0131] Furthermore, in this embodiment, referring to the relevant description of the feature extractor in the region recognition model, similarly, the feature extractor in the preset recognition model includes a first convolutional neural network, a convolutional-recurrent neural network, and a recurrent-fully convolutional neural network. The convolutional-recurrent neural network includes a second convolutional neural network and a first recurrent neural network. Since the second convolutional neural network can be a ResNet18 trained and fine-tuned based on Places365, and the image in Places365 is very similar to the image in the fourth material image, in order to save the processing resources of the graphics processing unit (GPU), reduce the training complexity of the preset recognition model, and avoid overfitting of the region recognition model, when training the model parameters of the preset recognition model, only the first recurrent neural network in the preset recognition model can be trained, while the model parameters of the second convolutional neural network in the preset recognition model are frozen. Therefore, this application provides a possible implementation method in which the feature extractor in the preset recognition model includes a first convolutional neural network, a convolutional-recurrent neural network, and a recurrent-fully convolutional neural network. The convolutional-recurrent neural network includes a second convolutional neural network and a first recurrent neural network. The method may also include, for example, S14: when training the model parameters of the preset recognition model, freezing the model parameters of the second convolutional neural network in the preset recognition model and training the first recurrent neural network in the preset recognition model.

[0132] In the process of training the preset recognition model to obtain the region recognition model, the optimizer used is stochastic gradient descent (SGD), with a momentum parameter set to 0.9. Additionally, a pre-warm-up learning rate strategy can be used, with a prediction coefficient of 0.0035 and 10 pre-warm-up rounds; after pre-warm-up, training continues with an initial learning rate of 0.1, using cosine annealing to decay the learning rate. Furthermore, early stopping can be used to avoid overfitting of the region recognition model.

[0133] Based on the above description of the region recognition model obtained by training the preset recognition model, the region recognition model can achieve an accuracy of 0.920 on the test set. For example, the Kappa coefficient used to measure the recognition accuracy of the region recognition model is 0.720, and the F1 index used to represent the evaluation index of the region recognition model is 0.773. Referring to the table below, the accuracy, Kappa coefficient, and F1 index of the region recognition model trained based on one or more of the following data: the first material image of the target unit, the second material image of the target unit, and the first time series of the target unit. The accuracy of the region recognition model trained based on either the first or second material image of the target unit can reach above 0.8. The accuracy of the region recognition model trained using both the first and second material images of the target unit shows a significant improvement. In particular, the accuracy of the region recognition model trained using all three data—the first, second, and first time series of the target unit—can reach 0.920, and the Kappa coefficient of this region recognition model can reach 0.720. The Kappa coefficient of this region recognition model can also reach 0.773, indicating the effectiveness of the fusion of the three data.

[0134] surface

[0135]

[0136]

[0137] Wherein, I1 represents the first material image of the unit to be identified, I2 represents the second material image of the unit to be identified, and T1 represents the first time series of the unit to be identified; CNN1 represents the first convolutional neural network, CNN2-RNN1 represents the convolutional-recurrent neural network combining the second convolutional neural network and the first recurrent neural network, and RNN2-FCN represents the recurrent-fully convolutional neural network combining the second recurrent neural network and the fully convolutional neural network.

[0138] See Figure 6 , Figure 6 This diagram illustrates the confidence level of a region recognition model that correctly identifies the region to which a sample unit belongs, as provided in an embodiment of this application. Only 20% of the sample units have a confidence level less than 0.7 for correctly identifying their region, and the average confidence level for all correctly identified sample units is 0.897. This indicates that the region recognition model successfully identifies the regional features of most sample units, resulting in a relatively high confidence level for most sample units. Similarly, see... Figure 7 , Figure 7This is a schematic diagram of the confidence distribution of the correct identification of the region to which a sample unit belongs, provided in an embodiment of this application; wherein, the confidence distribution of the correct identification of the region to which the sample unit belongs by the region identification model indicates that the region identification model has a high level of correct identification ability for most regions.

[0139] In addition, the training process of the above-mentioned preset recognition model was completed using Python 3.7 under the PyTorch 1.10.0 framework, and the GPU used could be, for example, an NVIDIA GeForce RTX 2080 Ti.

[0140] As can be seen from the above technical solution, firstly, the spatial material image of the space to be identified, which contains multiple objects, is divided into units from an overhead view to obtain the unit to be identified and the first material image of the unit to be identified; second material image of the unit to be identified from the object's perspective is obtained; and key point clustering is performed on the key point information in the object activity information of the unit to be identified according to a preset time period to obtain the first time series of the unit to be identified; based on the description of the material information of the unit to be identified from the overhead view in the first material image and the supplementary description of the material information of the unit to be identified from the object's perspective in the second material image, the first time series is added to further describe the object activity pattern of the unit to be identified, so as to further reflect the regional characteristics of the area to which the unit to be identified belongs. Then, the first material image, the second material image, and the first time series are input into the feature extractor of the region recognition model for feature extraction, resulting in first, second, and third depth features. The first fused feature, obtained by fusing the first, second, and third depth features, is then input into the classification layer of the region recognition model for region recognition, yielding the region recognition result for the unit to be identified. The first fused feature integrates the material features of the unit to be identified from a bird's-eye view, the material features from an object's perspective, and object activity features reflecting the regional characteristics of the region to which the unit belongs. This provides a more comprehensive description of the regional features of the unit to be identified, making the region recognition results more accurate. This method reduces region recognition errors and improves the accuracy of region recognition results even when different regions with similar material morphologies exist in space, thus meeting the high-precision requirements for region recognition in space.

[0141] Based on the foregoing description, when the spatial region identification method is applied to the target functional area identification scenario in urban space, taking the space to be identified as the city to be identified, the object as a vehicle, the spatial material image as a remote sensing image, the first material image as a remote sensing image patch, multiple local material images as multiple street view images, the second material image as a street view image, the object activity information as vehicle travel information, the key points as the travel start and end points, the preset time period as every 1 hour, and the region identification model as a target functional area identification model, the following is a detailed description of a target functional area identification method in urban space provided by the embodiments of this application. See also Figure 8 , Figure 8 A flowchart illustrating a method for identifying target functional areas in urban space, provided in this application embodiment, is included. This method can be executed by a server and includes the following steps:

[0142] S801: Divide the remote sensing image of the city to be identified from an overhead view into units to obtain the units to be identified and the remote sensing image blocks of the units to be identified in the city; the city to be identified has multiple vehicles clustered together.

[0143] S802: Connect multiple street view images of the city to be identified from the vehicle's perspective according to the unit to be identified, and obtain the street view image of the unit to be identified.

[0144] S803: Extract the origin and destination of vehicles in the city to be identified, respectively, to obtain the origin and destination information of the city to be identified.

[0145] S804: Connect the units according to the travel origin information and travel destination information of the city to be identified, and obtain the travel origin information and travel destination information of the unit to be identified.

[0146] S805: Perform trip origin clustering and trip destination clustering on the trip origin information and trip destination information of the unit to be identified every hour to obtain the first time series obtained by fusing the trip origin time series and trip destination time series of the unit to be identified.

[0147] S806: The first depth feature is obtained by extracting features from remote sensing image blocks through the first convolutional neural network in the target functional area recognition model.

[0148] S807: The second deep feature is obtained by extracting features from the street view image of the unit to be identified through the convolutional recurrent neural network in the target functional area recognition model.

[0149] S808: The third deep feature is obtained by extracting features from the first time series through the recurrent fully convolutional neural network in the target functional area recognition model.

[0150] S809: Through the classification layer in the target functional region identification model, the first fused feature obtained by feature fusion of the first deep feature, the second deep feature and the third deep feature is used to identify the target functional region of the unit to be identified.

[0151] Of course, the spatial area identification method provided in this application embodiment can also be applied to other functional area identification scenarios in urban space, or land use area identification scenarios in urban space, or other specific area identification scenarios in other spaces, etc.

[0152] It should be noted that, based on the implementation methods provided in the above aspects, this application can be further combined to provide more implementation methods.

[0153] based on Figure 2 In accordance with the spatial region identification method provided in the corresponding embodiments, this application also provides a spatial region identification device, see [link to relevant documentation]. Figure 9 , Figure 9 This is a structural diagram of a spatial region identification device provided in an embodiment of the present application. The spatial region identification device 900 includes: a division unit 901, an acquisition unit 902, a clustering unit 903, an extraction unit 904, and an identification unit 905.

[0154] The segmentation unit 901 is used to segment the spatial material image of the space to be identified from an overhead view to obtain the unit to be identified in the space to be identified and the first material image of the unit to be identified; the space to be identified contains multiple objects.

[0155] Acquisition unit 902 is used to acquire a second material image of the unit to be identified from the perspective of the object;

[0156] Clustering unit 903 is used to perform key point clustering on key point information in the object activity information of the unit to be identified according to a preset time period, so as to obtain the first time series of the unit to be identified.

[0157] The extraction unit 904 is used to extract features from the first material image, the second material image and the first time series respectively through the feature extractor in the region recognition model to obtain the first depth feature, the second depth feature and the third depth feature;

[0158] The recognition unit 905 is used to perform region recognition on the first fused feature obtained by feature fusion of the first deep feature, the second deep feature and the third deep feature through the classification layer in the region recognition model, and obtain the region recognition result of the unit to be recognized.

[0159] In one possible implementation, the acquisition unit 902 is specifically used for:

[0160] The second material image is obtained by connecting multiple local material images of the space to be identified from the object's perspective based on the unit to be identified.

[0161] In one possible implementation, clustering unit 903 is specifically used for:

[0162] Key points are extracted from the object activity information in the space to be identified to obtain key point information of the space to be identified;

[0163] Based on the key point information of the space to be identified, the units are connected to obtain the key point information of the units to be identified.

[0164] Based on a preset time period, key point clustering is performed on the key point information of the unit to be identified to obtain the first time series.

[0165] In one possible implementation, clustering unit 903 is specifically used for:

[0166] Based on the object activity information of the space to be identified, the units are connected to obtain the object activity information of the unit to be identified.

[0167] Key points are extracted from the object activity information of the unit to be identified to obtain the key point information of the unit to be identified.

[0168] Based on a preset time period, key point clustering is performed on the key point information of the unit to be identified to obtain the first time series.

[0169] In one possible implementation, the device further includes: a normalization unit and an update unit;

[0170] The normalization unit is used to normalize the first time series to obtain the normalized first time series.

[0171] The update unit is used to update the first time series based on the normalized first time series.

[0172] In one possible implementation, the feature extractor in the region recognition model includes a first convolutional neural network, a convolutional-recurrent neural network, and a recurrent-fully convolutional neural network;

[0173] Extraction unit 904 is specifically used for:

[0174] The first convolutional neural network in the region recognition model is used to extract features from the first material image to obtain the first depth features;

[0175] The second material image is used to extract features by using a convolutional-recurrent neural network in the region recognition model to obtain the second depth features.

[0176] By using a recurrent fully convolutional neural network in the region recognition model, features are extracted from the first time series to obtain the third deep features.

[0177] In one possible implementation, the convolutional-recurrent neural network includes a second convolutional neural network and a first recurrent neural network;

[0178] Extraction unit 904 is specifically used for:

[0179] The second convolutional neural network in the region recognition model is used to extract features from the second material image to obtain the first convolutional features;

[0180] The first recurrent neural network in the region recognition model is used to process the first convolutional features to obtain the second deep features.

[0181] In one possible implementation, the recurrent-fully convolutional neural network includes a second recurrent neural network and a fully convolutional neural network;

[0182] Extraction unit 904 is specifically used for:

[0183] The second recurrent neural network in the region recognition model is used to extract features from the first time series to obtain the second convolutional features.

[0184] The first time series is used to extract features by using a fully convolutional neural network in the region recognition model to obtain time features;

[0185] The second convolutional feature and the temporal feature are fused to obtain the third deep feature.

[0186] In one possible implementation, the device further includes: a training unit;

[0187] The training unit is specifically used for:

[0188] The spatial material image of the sample space from an overhead view is divided into units to obtain sample units and third material images of the sample units; each sample unit has regional label data that identifies the region to which it belongs.

[0189] Acquire a fourth material image of the sample unit from the object's perspective;

[0190] Based on a preset time period, key point clustering is performed on the key point information in the object activity information of the sample unit to obtain the second time series of the sample unit.

[0191] By using the feature extractor in the preset recognition model, features are extracted from the third material image, the fourth material image, and the second time series to obtain the fourth deep feature, the fifth deep feature, and the sixth deep feature.

[0192] By using the classification layer in the preset recognition model, the second fusion feature obtained by feature fusion of the fourth, fifth and sixth deep features is used to identify regions and obtain the region prediction results of sample units.

[0193] Based on the region label data, region prediction results, and the loss function of the preset recognition model, the model parameters of the preset recognition model are trained to obtain the region recognition model.

[0194] In one possible implementation, when the difference in the number of sample units in different regions is greater than a preset difference value, the loss function is specifically a weighted cross-entropy loss function, in which the weights are used to balance the difference in the number of units.

[0195] In one possible implementation, the feature extractor in the preset recognition model includes a first convolutional neural network, a convolutional-recurrent neural network, and a recurrent-fully convolutional neural network, the convolutional-recurrent neural network includes a second convolutional neural network and a first recurrent neural network, and the device further includes: a freezing unit;

[0196] The freeze unit is used to freeze the model parameters of the second convolutional neural network in the preset recognition model and train the first recurrent neural network in the preset recognition model when training the model parameters of the preset recognition model.

[0197] As can be seen from the above technical solution, firstly, the spatial material image of the space to be identified, which contains multiple objects, is divided into units from an overhead view to obtain the unit to be identified and the first material image of the unit to be identified; second material image of the unit to be identified from the object's perspective is obtained; and key point clustering is performed on the key point information in the object activity information of the unit to be identified according to a preset time period to obtain the first time series of the unit to be identified; based on the description of the material information of the unit to be identified from the overhead view in the first material image and the supplementary description of the material information of the unit to be identified from the object's perspective in the second material image, the first time series is added to further describe the object activity pattern of the unit to be identified, so as to further reflect the regional characteristics of the area to which the unit to be identified belongs. Then, the first material image, the second material image, and the first time series are input into the feature extractor of the region recognition model for feature extraction, resulting in first, second, and third depth features. The first fused feature, obtained by fusing the first, second, and third depth features, is then input into the classification layer of the region recognition model for region recognition, yielding the region recognition result for the unit to be identified. The first fused feature integrates the material features of the unit to be identified from a bird's-eye view, the material features from an object's perspective, and object activity features reflecting the regional characteristics of the region to which the unit belongs. This provides a more comprehensive description of the regional features of the unit to be identified, making the region recognition results more accurate. This method reduces region recognition errors and improves the accuracy of region recognition results even when different regions with similar material morphologies exist in space, thus meeting the high-precision requirements for region recognition in space.

[0198] This application also provides a computer device, which may be a server, see [link to previous document]. Figure 10 , Figure 10 This application provides a structural diagram of a server. The server 1000 can vary significantly due to different configurations or performance. It may include one or more processors, such as a Central Processing Unit (CPU) 1022, and a memory 1032, as well as one or more storage media 1030 (e.g., one or more mass storage devices) for storing application programs 1042 or data 1044. The memory 1032 and storage media 1030 can be temporary or persistent storage. The program stored in the storage media 1030 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the server. Furthermore, the CPU 1022 may be configured to communicate with the storage media 1030 and execute the series of instruction operations in the storage media 1030 on the server 1000.

[0199] Server 1000 may also include one or more power supplies 1026, one or more wired or wireless network interfaces 1050, one or more input / output interfaces 1058, and / or one or more operating systems 1041, such as Windows Server. TM Mac OS X TM Unix TM Linux TM FreeBSD TM etc.

[0200] In this embodiment, the central processing unit 1022 in server 1000 can perform the following steps:

[0201] The spatial material image of the space to be identified from an overhead view is divided into units to obtain the unit to be identified and the first material image of the unit to be identified in the space to be identified; the space to be identified contains multiple objects.

[0202] Acquire a second material image of the unit to be identified from the object's perspective;

[0203] Based on a preset time period, key point clustering is performed on the key point information in the object activity information of the unit to be identified to obtain the first time series of the unit to be identified.

[0204] The feature extractor in the region recognition model is used to extract features from the first material image, the second material image and the first time series respectively, to obtain the first deep feature, the second deep feature and the third deep feature.

[0205] By using the classification layer in the region recognition model, the first fused feature obtained by feature fusion of the first deep feature, the second deep feature, and the third deep feature is used to identify the region, and the region recognition result of the unit to be identified is obtained.

[0206] The computer device provided in this application embodiment can also be a terminal, see [link to relevant documentation]. Figure 11 , Figure 11This is a structural diagram of a terminal provided in an embodiment of this application. Taking a smartphone as an example, the smartphone includes: a radio frequency (RF) circuit 1110, a memory 1120, an input unit 1130, a display unit 1140, a sensor 1150, an audio circuit 1160, a wireless Fidelity (WiFi) module 1170, a processor 1180, and a power supply 1190, etc. The input unit 1130 may include a touch panel 1131 and other input devices 1132, the display unit 1140 may include a display panel 1141, and the audio circuit 1160 may include a speaker 1161 and a microphone 1162. Those skilled in the art will understand that... Figure 11 The smartphone structure shown does not constitute a limitation on smartphones and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0207] The memory 1120 can be used to store software programs and modules. The processor 1180 executes various functions and data processing of the smartphone by running the software programs and modules stored in the memory 1120. The memory 1120 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the smartphone (such as audio data, phonebook, etc.). In addition, the memory 1120 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0208] The processor 1180 is the control center of the smartphone, connecting various parts of the smartphone via various interfaces and lines. It performs various functions and processes data by running or executing software programs and / or modules stored in the memory 1120 and calling data stored in the memory 1120. Optionally, the processor 1180 may include one or more processing units; preferably, the processor 1180 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1180.

[0209] In this embodiment, the processor 1180 in the smartphone can perform the following steps:

[0210] The spatial material image of the space to be identified from an overhead view is divided into units to obtain the unit to be identified and the first material image of the unit to be identified in the space to be identified; the space to be identified contains multiple objects.

[0211] Acquire a second material image of the unit to be identified from the object's perspective;

[0212] Based on a preset time period, key point clustering is performed on the key point information in the object activity information of the unit to be identified to obtain the first time series of the unit to be identified.

[0213] The feature extractor in the region recognition model is used to extract features from the first material image, the second material image and the first time series respectively, to obtain the first deep feature, the second deep feature and the third deep feature.

[0214] By using the classification layer in the region recognition model, the first fused feature obtained by feature fusion of the first deep feature, the second deep feature, and the third deep feature is used to identify the region, and the region recognition result of the unit to be identified is obtained.

[0215] According to one aspect of this application, a computer-readable storage medium is provided for storing a computer program for implementing the spatial region identification method described in the foregoing embodiments.

[0216] According to one aspect of this application, a computer program product is provided, comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the methods provided in various optional implementations of the above embodiments.

[0217] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0218] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0219] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0220] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0221] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0222] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing computer programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0223] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for identifying regions in space, characterized in that, The method includes: The spatial material image of the space to be identified, viewed from an overhead perspective, is divided into units to obtain the unit to be identified in the space and the first material image of the unit to be identified; the space to be identified contains multiple objects. Acquire a second material image of the unit to be identified from the object's perspective; According to a preset time period, key point clustering is performed on the key point information corresponding to the unit to be identified in the object activity information of the space to be identified, so as to obtain the first time series of the unit to be identified. The first time series is used to describe the object activity pattern of the unit to be identified, so as to reflect the regional characteristics of the area to which the unit to be identified belongs. The feature extractor in the region recognition model is used to extract features from the first material image, the second material image and the first time series respectively to obtain the first deep feature, the second deep feature and the third deep feature. The region identification result of the unit to be identified is obtained by performing region identification on the first fused feature obtained by feature fusion of the first depth feature, the second depth feature and the third depth feature through the classification layer in the region identification model.

2. The method according to claim 1, characterized in that, The specific steps of obtaining the second material image of the unit to be identified from the object's perspective are as follows: The second material image is obtained by connecting multiple local material images of the space to be identified from the object's perspective according to the unit to be identified.

3. The method according to claim 1, characterized in that, The step of performing key point clustering on key point information in the object activity information of the unit to be identified according to a preset time period to obtain the first time series of the unit to be identified includes: Key points are extracted from the object activity information in the space to be identified to obtain key point information of the space to be identified. Based on the key point information of the space to be identified, the key point information of the unit to be identified is obtained by connecting the key point information of the unit to be identified. The key point information of the unit to be identified is clustered according to the preset time period to obtain the first time series.

4. The method according to claim 1, characterized in that, The step of performing key point clustering on key point information in the object activity information of the unit to be identified according to a preset time period to obtain the first time series of the unit to be identified includes: Based on the object activity information of the space to be identified, the object activity information of the unit to be identified is obtained by connecting the units to be identified; Key points are extracted from the object activity information of the unit to be identified to obtain key point information of the unit to be identified. The key point information of the unit to be identified is clustered according to the preset time period to obtain the first time series.

5. The method according to any one of claims 1-4, characterized in that, The method further includes: The first time series is normalized to obtain the normalized first time series; Update the first time series based on the normalized first time series.

6. The method according to claim 1, characterized in that, The feature extractor in the region recognition model includes a first convolutional neural network, a convolutional-recurrent neural network, and a recurrent-fully convolutional neural network; the feature extractor in the region recognition model extracts features from the first material image, the second material image, and the first time series to obtain first deep features, second deep features, and third deep features, including: The first depth feature is obtained by extracting features from the first material image through the first convolutional neural network in the region recognition model. The second material image is used to extract features by using the convolutional-recurrent neural network in the region recognition model to obtain the second depth feature. The third deep feature is obtained by extracting features from the first time series using the recurrent fully convolutional neural network in the region recognition model.

7. The method according to claim 6, characterized in that, The convolutional-recurrent neural network includes a second convolutional neural network and a first recurrent neural network; the step of extracting features from the second material image using the convolutional-recurrent neural network in the region recognition model to obtain the second depth features includes: The second material image is used to extract features through the second convolutional neural network in the region recognition model to obtain the first convolutional features. The second depth feature is obtained by processing the first convolutional feature through the first recurrent neural network in the region recognition model.

8. The method according to claim 6, characterized in that, The recurrent-fully convolutional neural network includes a second recurrent neural network and a fully convolutional neural network; the step of extracting features from the first time series using the recurrent-fully convolutional neural network in the region recognition model to obtain the third deep features includes: The second recurrent neural network in the region recognition model is used to extract features from the first time series to obtain the second convolutional features. The first time series is used to extract time features by using the fully convolutional neural network in the region recognition model; The second convolutional feature and the temporal feature are fused to obtain the third depth feature.

9. The method according to claim 1, characterized in that, The training steps of the region recognition model include: The spatial material image of the sample space from an overhead view is divided into units to obtain sample units in the sample space and third material images of the sample units; each sample unit has region label data that identifies the region to which the sample unit belongs; Obtain a fourth material image of the sample unit from the perspective of the object; Based on a preset time period, key point clustering is performed on the key point information in the object activity information of the sample unit to obtain the second time series of the sample unit. By using the feature extractor in the preset recognition model, features are extracted from the third material image, the fourth material image, and the second time series to obtain the fourth deep feature, the fifth deep feature, and the sixth deep feature. By using the classification layer in the preset recognition model, the second fused feature obtained by feature fusion of the fourth deep feature, the fifth deep feature, and the sixth deep feature is used to identify regions and obtain the region prediction result of the sample unit. The region identification model is obtained by training the model parameters of the preset identification model based on the region label data, the region prediction results, and the loss function of the preset identification model.

10. The method according to claim 9, characterized in that, When the difference in the number of sample units in different regions is greater than a preset difference value, the loss function is specifically a weighted cross-entropy loss function, in which the weights are used to balance the difference in the number of units.

11. The method according to claim 9, characterized in that, The feature extractor in the preset recognition model includes a first convolutional neural network, a convolutional-recurrent neural network, and a recurrent-fully convolutional neural network. The convolutional-recurrent neural network includes a second convolutional neural network and a first recurrent neural network. The method further includes: When training the model parameters of the preset recognition model, the model parameters of the second convolutional neural network in the preset recognition model are frozen, and the first recurrent neural network in the preset recognition model is trained.

12. A spatial region identification device, characterized in that, The device includes: a segmentation unit, an acquisition unit, a clustering unit, an extraction unit, and an identification unit; The segmentation unit is used to divide the spatial material image of the space to be identified into units from an overhead view, thereby obtaining the unit to be identified in the space to be identified and the first material image of the unit to be identified; the space to be identified contains multiple objects. The acquisition unit is used to acquire a second material image of the unit to be identified from the perspective of the object; The clustering unit is used to perform key point clustering on the key point information corresponding to the target unit in the object activity information of the target space according to a preset time period, so as to obtain the first time series of the target unit. The first time series is used to describe the object activity pattern of the target unit, so as to reflect the regional characteristics of the area to which the target unit belongs. The extraction unit is used to extract features from the first material image, the second material image, and the first time series respectively through the feature extractor in the region recognition model to obtain a first depth feature, a second depth feature, and a third depth feature. The recognition unit is used to perform region recognition on the first fused feature obtained by feature fusion of the first depth feature, the second depth feature and the third depth feature through the classification layer in the region recognition model, and obtain the region recognition result of the unit to be recognized.

13. The apparatus according to claim 12, characterized in that, The acquisition unit is specifically used for: The second material image is obtained by connecting multiple local material images of the space to be identified from the object's perspective according to the unit to be identified.

14. The apparatus according to claim 12, characterized in that, The clustering unit is specifically used for: Key points are extracted from the object activity information in the space to be identified to obtain key point information of the space to be identified. Based on the key point information of the space to be identified, the key point information of the unit to be identified is obtained by connecting the key point information of the unit to be identified. The key point information of the unit to be identified is clustered according to the preset time period to obtain the first time series.

15. The apparatus according to claim 12, characterized in that, The clustering unit is specifically used for: Based on the object activity information of the space to be identified, the object activity information of the unit to be identified is obtained by connecting the units to be identified; Key points are extracted from the object activity information of the unit to be identified to obtain key point information of the unit to be identified. The key point information of the unit to be identified is clustered according to the preset time period to obtain the first time series.

16. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store computer programs and to transfer the computer programs to the processor; The processor is configured to execute the method according to any one of claims 1-11 according to instructions in the computer program.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program for implementing the method according to any one of claims 1-11.

18. A computer program product, comprising a computer program, characterized in that, When it is run on a computer device, it causes the computer device to perform the method according to any one of claims 1-11.

Images